Teaching the pronunciation of Japanese double-mora phonemes using speech recognition technology

A CALL (computer-aided language learning) system was developed for teaching the pronunciation of Japanese double-mora phonemes to non-native speakers of Japanese. Double-mora vowels and single-mora vowels are spectrally almos t identical but their phone durations differ significantly. Similar conditi ons exist between moraic nasals and non-moraic nasals, and between moraic a nd non-moraic obstruents, Our CALL system asks the learner to read minimal pairs. Speech recognition technology is used to measure the durations of ea ch phone and the system tells the learner the likelihood of native speakers understanding the learner's utterance as the learner intended, These intel ligibility scores are based on perception experiments where native speakers judged the confusability of minimal pairs containing phones with various s ynthesized durations. The system then instructs the learner to either short en or lengthen his pronunciation. The learner can terminate training when h is communicative performance has met his expectations. For instance, when a learner hits a learning plateau, intelligibility indices can help him deci de whether further learning effort is worthwhile. Given that most adult lea rners can never attain complete nativeness, it is of practical use to be to ld when non-native accents cannot be removed further. Learning experiments show that learners quickly capture the relevant duration cues. (C) 2000 Els evier Science B.V. All rights reserved.